Publish without bias or perish without replications

Abstract

There is mounting evidence that a large portion of experimental results cannot be replicated, leading many to believe that science is now in the throes of a replicability crisis. In response, there have been calls to reduce publication bias against negative results because of the effect that publication bias has on the publication record. Others, however, argue that publication bias need not be detrimental to scientific progress. Here, we propose a novel mechanism by dint of which reducing publication bias can benefit science regardless of the effect that publication bias has on the publication record. To do so, we introduce a series of increasingly complex mathematical models. Our models represent a scientific community consisting of discovery researchers who test novel hypotheses, and confirmation researchers who test known hypotheses. Results show that reducing publication bias can have the surprising consequence of increasing the share of confirmation researchers who conduct replications. When a large share of scientists conduct confirmation research, scientists have an incentive to conduct high-quality research as others are likely to check their findings. Our models therefore suggest an underappreciated reasons why reducing publication bias might benefit science.